Real-time Road Surface LiDAR SLAM based on Dead-Reckoning Assisted Registration

In the advancement of vehicle detection sensors, LiDAR (Light Detection and Ranging) sensors are increasingly utilized for estimating the surrounding environment and vehicle status. Beyond traditional uses, recognizing ground shapes with LiDAR can become essential, enabling vehicles to adjust speed and suspension stiffness based on road surface conditions, thus reducing vibration and enhancing durability and stability. However, the use of mass-produced LiDAR presents challenges in accurately recognizing the shape of the ground due to sparse point clouds and limited field of view, necessitating advanced mapping techniques. Traditional LiDAR mapping systems often encounter issues with accuracy and stability due to the degradation of point cloud registration performance, which is affected by dynamic objects on the road. This research emphasizes the need for a swift, precise, and stable LiDAR mapping system, particularly in high-speed and urban environments. We propose an algorithm that effectively removes points from dynamic objects in real-time through a two-stage road surface extraction process, improving the accuracy and stability of point cloud registration in LiDAR odometry by optimizing variable reduction and integrating dead reckoning with in-vehicle sensors. This approach allows for more accurate and stable road surface mapping around the vehicle, even in the presence of numerous dynamic objects and with mass-produced LiDARs. Our algorithm, tested in various terrains and both urban and highway settings, showed significant improvements in mapping accuracy and computational speed, approximately 16% and 45%, respectively, over current state-of-the-art methods, as validated through quantitative comparisons with high-precision mapping equipment.